Littera Deusto

Modern Languages, Basque Studies and Humanities

Speech recognition

marzo 22nd, 2009 · No hay Comentarios

Speech recognition converts spoken sounds (captured by a microphone) to readable words. It should not be confused with speaker recognition, which consists in identifying the person speaking by analyzing their voice.

There are two types of speech recognition systems: discrete and continuous. The first requires that the separation between each word be marked clearly, whereas the second understands natural talking patterns. It has only been a few years since most softwares have started to distinguish continuous speech.

To avoid mixing up similar-sounding words, old approaches tried applying grammatical and syntactical rules. This restricted the number of possible choices, but difference accents, dialects and other irregularities still presented a lot of problems. Current systems turn to statistical models based on large speech corpora to determine the most appropriate answer.

Speech recognition has many purposes, among them allowing the physically disabled to express themselves through dictation and navigate their computers using their voice. Another interesting assistance would be to court reporting, where testimonies could be written down real-time.

References:

Etiquetas:

  • Etiquetas